You are on page 1of 8

Electric Power Systems Research 76 (2006) 485–492

Maximum loadability of power systems using


hybrid particle swarm optimization
Amgad A. EL-Dib ∗ , Hosam K.M. Youssef 1 , M.M. EL-Metwally 1 , Z. Osman 1
Electrical Power and Machines Department, Faculty of Engineering, Cairo University, Giza, Egypt
Received 16 December 2004; accepted 15 May 2005
Available online 27 December 2005

Abstract
The problem of voltage stability is one of the main concerns in the operation of power systems. There are different approaches to estimate the
voltage stability of the system. One of these approaches is to find the margin from the current operating point to the maximum loading point of the
system. Finding this maximum loading point can be formulated as an optimization problem. This paper utilizes the newly developed evolutionary
particle swarm optimization in solving this optimization problem. Details of the implementation of the proposed method to two test systems
(Ward-Hale 6-bus) and (IEEE 14-bus) are presented. The results are compared to those obtained by the widely used continuation power flow
technique. Good agreement has been obtained proving the validity and applicability of the proposed method.
© 2005 Published by Elsevier B.V.

Keywords: Power system optimization; Maximum loadability; Voltage stability; Hybrid particle swarm optimization

1. Introduction are static since the problem of voltage stability can be seen as a
static one although it is dynamic in its nature. Since one of the
The complexity of power systems is continuously growing main aspects causing the voltage collapse is the slow increase
due to the increasing number of generation plants and load of load with respect to other parameters in the network and with
demand. Power systems are becoming heavily stressed due to sufficient damping [1], voltage stability can be seen as a static
the increased loading of the transmission lines and due to the problem rather than a dynamic one. This way the problem can
difficulty of constructing new transmission systems as well as be analyzed in a much simpler way using only algebraic equa-
the difficulty of building new generating plants near the load tions. The different approaches used in the literature include the
centers. All of these problems lead to the voltage stability prob- following: trying to find some indicator which shows how far the
lem in the system. There have been a number of incidents in system is from the maximum loading point like the determinant
the past years which were diagnosed as voltage stability prob- or the smallest eigenvalue or singular value of the system Jaco-
lems due to increased loading and decreased stability margin. bian [2–4]. Another indicator is based on the fact that at each
The stability margin may be defined as the distance between loading on the system, there are two solutions for the power flow
the loading of the system and the maximum loading limit of problem. As the loading increases these two solutions get closer
the system. As this margin decreases the system approaches a to each other. Therefore, the closeness of the second power flow
very critical stage since if one of the contingencies occurs, this solution was used as an indicator of how close the system to the
may lead the system to a total collapse. Therefore, in recent maximum loading point. The advantage of these indicators is
years there have been a lot of research in how to determine the that no loading pattern or scenario has to be defined. However,
voltage stability of the system or in other words to determine in fact any system faces a load increase with a certain pattern
the margin between the operating point of the system and the according to its load forecasting data, therefore, such indices
maximum loading point. Most of these methods or indicators may not provide satisfactory information regarding the close-
ness to the critical point. Also these indices remain relatively
stable up to the system stability limit where they suddenly and
∗ Corresponding author. rapidly change. This type of behavior renders these indices inad-
1 Member, IEEE. equate to detect stability problems, since they give a false sense

0378-7796/$ – see front matter © 2005 Published by Elsevier B.V.


doi:10.1016/j.epsr.2005.05.011
486 A.A. EL-Dib et al. / Electric Power Systems Research 76 (2006) 485–492

of security by remaining practically unchanged until the system IEEE 14-bus test systems. In Section 5, the results are analyzed.
is too close to the stability point [5]. Another technique is the Finally, the conclusions are given in Section 6.
repetitive power flow solution by increasing the load on the sys-
tem in some direction in steps and solving the load flow at each 2. Formulation of maximum loading point as an
step until the load flow solution diverges [1,6]. At this point, it optimization problem
was assumed that the system is at its maximum loading point,
however this technique has some disadvantages like: The problem of finding the maximum loading of a power
system can be formulated as an optimization problem as in [6].
• The point where power flow calculation diverges (which is a A loading factor is used which represents the increase in the
mathematical failure) does not represent the maximum load- system load from some initial operating point. The optimization
ing point (which is a physical limitation). Therefore, the diver- function or the cost function is to maximize this factor or to min-
gence of the load flow may be due to mathematical problem in imize the reciprocal of this factor while satisfying the following
the utilized method. This means that the point of divergence constraints:
is not always the maximum loading point [6].
• It has also been noticed that the point where the calculations 1. Power flow equations (power balance equations at each bus
diverge may vary depending on the method which was used for both active and reactive powers) [12].
[1]. 2. A certain direction or scenario for this load increase.
3. Other constraints like the reactive power capability of the
The divergence problem of the power flow techniques was PV-buses.
solved by the continuation power flow (CPF) [7] which depends
on reformulating the power flow equations and applying a locally These are the basic constraints so that the formulation will actu-
parameterized continuation technique. It is basically a mathe- ally model the physical meaning of the problem.
matical method used to find a path of equilibrium solutions of Generally, any optimization problem can be formulated as:
a set of nonlinear equations. This technique has a wide range
of applications like evaluation of the maximum loading point min . f (x) (1)
as well as simulation of the power system static behavior due to
s.t. g(x) = 0 (2)
load and/or generation variation with or without control devices.
However, there are some difficulties in the CPF like the complex x1 ≤ x ≤ xu (3)
mathematical formulations, or if the system is already near the
maximum loading limit the CPFLOW may face some conver- where x is the vector of the problem variables, f(x) the objective
gence problems. function (cost function), g(x) are the equality constraints of the
Another technique has been adopted in order to find the max- problem and finally Eq. (3) represents the inequality constraints
imum loading point of the system in [6], where the problem was included in the problem.
formulated as an optimization problem whose objective was to A mathematical formulation for the problem may be as fol-
maximize the load on the system in some direction while sat- lows:
isfying some constraints. The most important constraint is the First, the normal power flow equations have to be refor-
power flow equations with some other operational constraints. mulated to include the loading factor and the direction of this
This problem was solved using a numerical technique which is increase.
the Sequential Quadratic Programming (SQP) algorithm or by
using the Interior Point (IP) optimization method [8]. The SQP Let Vi : voltage magnitude of bus i;
method includes the differentiation of the constraints and also Vj : voltage magnitude of bus j;
the formation of a number of matricies in each iteration and δi : voltage angle of bus i;
matrix inversion. δj : voltage angle of bus j;
In this paper, a new solution technique is proposed to solve Yij : magnitude of Yij element in bus admittance matrix;
the maximum loading point problem. This solution is based θ ij : angle of Yij element in bus admittance matrix;
on the newly developed evolutionary technique, the particle Pgi : generated power at bus i;
swarm optimization (PSO). PSO is an optimization technique Pli : load power at bus i;
formulated by Kennedy and Eberhart [9] inspired by the nat- Pti : net power injected at bus i;
ural behavior of a population of birds or insects. It simulates Qgi : generated reactive power at bus i;
their performance during the search for food. The PSO has been Qli : load reactive power at bus i;
utilized in some applications in the power system field such as Qti : net reactive power injected at bus i;
the solution of OPF, Reactive Power Planning or the Economic N: number of buses in the system;
Dispatch problems [10,11]. This paper is organized as follows: Npq : set of PQ-buses;
Section 2 describes the formulation of the maximum loading Npv : set of PV-buses.
point problem as an optimization problem. In Section 3, a back-
ground on the PSO technique is introduced. Section 4 describes The power flow equations, at any given bus i in the system
the application of the proposed method on the Ward-Hale 6- and are as follows:
A.A. EL-Dib et al. / Electric Power Systems Research 76 (2006) 485–492 487

Active power: are included. If the reactive power output of the generator is
within the limits then the voltage of this bus is maintained at its
Pti = Pgi − Pli (4) scheduled value, but if the reactive power output is not within
Pti − Pgi + Pli = 0 (5) the limits then the reactive power output is set at the violated
 limit and the voltage of this bus is free to change.
Pti = Vi Vj Yij cos(θij + δi − δj ) (6) The problem of maximizing the load of the system can then
j
 be stated as follows:
Vi Vj Yij cos(θij + δi − δj ) − Pgi + Pli = 0 (7) 1
j min . (16)
λ
Reactive power: s.t. g(x, λ) = 0 (17)
Qti = Qgi − Qli (8) z1 < z < zu (18)
Qti − Qgi + Qli = 0 (9) where λ is the loading factor; x = [δ V]t for all buses except the
 slack bus; g(x, λ) the set of reformulated power flow equations;
Qti = Vi Vj Yij sin(θij + δi − δj ) (10)
j z the vector of all variables which may be limited by some fixed
 values.
Vi Vj Yij sin(θij + δi − δj ) − Qgi + Qli = 0 (11)
j
3. Hybrid particle swarm optimization (HPSO)
where j = 1:N.
Let fpi and fqi be two functions defined as follows: The particle swarm optimization algorithm was originally
 introduced by Kennedy and Eberhart in 1995 as an alternative to
fpi = Vi Vj Yij cos(θij + δi − δj ) − Pgi + Pli (12) the standard Genetic algorithm (GA) [9]. The PSO was inspired
j
 by insect swarms and has since proven to be a competitor to the
fqi = Vi Vj Yij sin(θij + δi − δj ) − Qgi + Qli (13) GA when it comes to function optimization. The PSO model
j
consists of a number of particles moving around in the search
Now these equations should be reformulated as said before: space, each representing a possible solution to a numerical prob-
lem. Each particle has a position vector (xi ) and a velocity vector
Let λ: loading factor; (vi ). The position (pbesti ) is the best position encountered by the
kli : multiplier to designate the rate of load change at bus i as λ particle (i) during its search and the position (gbest) is that of
changes; the best particle in the swarm group.
ϕi : power factor angle of load change at bus i. so the active In each iteration, the velocity of each particle is updated
power equation will be: according to its best-encountered position and the best position
encountered among the group, using the following equation:

fpi = Vi Vj Yij cos(θij + δi − δj ) − Pgi + Pli + λkli cos ϕi vi = χ[wvi + a1i (pbesti − xi ) + a2i (gbest − xi )] (19)
j
(14) where χ is known as the constriction coefficient; w the inertia
weight; a1 , a2 are the random values different for each particle
This added term describes the increase in the active power and for each dimension between 0 and 2 [13].
load at bus i. The position of each particle is then updated in each iteration
The same is done with the reactive power equation: by adding the velocity vector to the position vector.
 x i = x i + vi (20)
fqi = Vi Vj Yij sin(θij + δi − δj ) − Qgi + Ql + λkli sin ϕi
j
(15) Eq. (19) consists of three terms: the first one depends on the
particle’s previous speed, the second term depends on the dis-
tance between the particle’s best previous and current position.
This added term gives the ability to increase the load at any Finally, the effect of the swarm group best experience on the
combination of buses and with any desired power factor which velocity of each individual in the group. This effect is consid-
is very suitable to the practical application. ered in Eq. (19) as the distance between swarms’ best experience
In this work, the entire increased load has been carried by (the position of the best particle in the swarm group) and the ith
the slack bus, however, a term can be added to the equations to particle’s current position. Eq. (20) simulates the flying of the
let other generators share in the increased load. Reformulating particle toward a new position. The role of the inertia weight w
the power flow equations is very important to model the sys- is considered very important in PSO convergence behavior [13].
tem actually at the maximum loading point. These power flow The inertia weight is employed to control the impact of the pre-
equations represent the equality constraints in the optimization vious history of velocities on the current velocity. In this way, the
problem. parameter w regulates the trade-off between the global and local
The next set of constraints is the inequality constraints. In exploration abilities of the swarm. A large inertia weight facili-
this paper, only the reactive power capabilities of the PV-buses tates global exploration (searching new areas), while a small one
488 A.A. EL-Dib et al. / Electric Power Systems Research 76 (2006) 485–492

tends to facilitate local exploration, i.e. fine-tuning the current The most important constraint in the process of finding the
search area. A suitable value for the inertia weight w usually pro- maximum loading point of the system is the reactive power
vides balance between global and local exploration abilities and capability of the PV-buses. Therefore, it is essential to model
consequently a reduction on the number of iterations required the PV-buses correctly. The voltage at a PV-bus should be kept
to locate the optimum solution. There has been a lot of research constant at its scheduled value (obtained from optimal power
in how to improve the performance of the PSO with respect to flow) until the maximum reactive power output of this bus is
the speed of convergence and to make sure that the PSO will reached. Then the reactive power should be kept constant at this
not get stuck in a local minima [14–15]. The improvements in maximum limit and the voltage will drop under the scheduled
the PSO are done by trying to have some of the properties as in value.
the GA beside the PSO own properties. One of the most pow- To include the inequality constraints of the reactive power
erful properties of the GA is the ability to breed and produce capability of the generators, the following procedure was
better individuals (children) than the old ones (parents). This applied:
technique is used in the algorithm proposed by this work. It is
used to accelerate the solution of the problem. • If Qminq < Qpvq < Qmaxq
A hybrid model of the standard GA and the PSO is intro- The algorithm should try to fix the voltage of the PV-bus
duced in [16]. This model incorporates one major aspect of the at its scheduled value. This way the inequality constraint is
standard GA into the PSO, which is the reproduction or breed- transformed to an equality constraint.
ing. Breeding is one of the core elements that make the standard • If Qpvq > Qmaxq or Qpvq < Qminq
GA a powerful algorithm. Therefore, a hybrid PSO with the The equality constraint will be to fix the reactive power
breeding property has the potential to reach a better optimum output to the violated limit and the voltage magnitude of this
than the standard PSO. The model for the breeding process is as bus will be free to change, where q = 1:Npv .
mentioned in [16]:
For the positions of the children: After defining the constraints, the objective function to be min-
imized by HPSO is:
child1 (xi ) = pi × parent1 (xi ) + (1 − pi ) × parent2 (xi ) (21)
  
1 2
child2 (xi ) = pi × parent2 (xi ) + (1 − pi ) × parent1 (xi ) (22) f (x, λ) = w1 + w2 [ (f 2 + fqi2 )]
λ i pi
For the velocity vectors of the children: 
+ w3 [ ((Qpvq − Qmax q )2 )Qviolmax
q
(parent1 (v) + parent2 (v)) × |parent1 (v)|
child1 (v) = (23)
|parent1 (v) + parent2 (v)| +((Qpvq − Qmin q )2 )Qviol
max

(parent1 (v) + parent2 (v)) × |parent2 (v)| +((Vpvq − Qschq )2 )V viol ] (26)
child2 (v) = (24)
|parent1 (v) + parent2 (v)|
where w1 , w2 , w3 are the weighting factors; Vsch, the scheduled
where pi is a uniformly distributed random number between PV-buses voltages; Qmax , vector of maximum reactive power
[0,1]; parent1 (xi ), the position vector of a randomly chosen par- generation at PV-buses; Qmin , vector of minimum reactive power
ticle to take part in the breeding process; parent2 (xi ), the position generation at PV-buses; i.e. set of all buses except slack bus;
vector of a randomly chosen particle to be the other parent in Qviol
max , binary vector of length equal to the number of PV-buses.
the breeding process; child1 (xi ), the position vector of the first Each element takes the value 1 or 0. It takes the value 1, if the
offspring; child2 (xi ), the position vector of the second offspring; bus corresponding to this element has violated the max. reactive
parent1 (v), the velocity vector of the first parent; parent2 (v) is power Qmaxq or its voltage is lower than its scheduled value;
the velocity vector of the second parent. Qviol
min , as Qmax but with the min. reactive power limit. However,
viol

usually when searching for the maximum loading point the con-
4. Application of HPSO straint to be violated is the constraint with the maximum reactive
power limit; Vviol , binary vector whose elements take the value
The HPSO tries to find the minimum of objective function by 1 if the respective elements in the above two vectors are zero
the principle stated before. First, we have to define our objective otherwise it will take the value 0.
function to satisfy all requirements of the optimization problem. Using this formulation, for every one of the PV-buses only
To include the equality constraints in the problem, the Penalty one element in the three vectors will take the value 1 and the
method is applied. The basic objective function of the problem other two will be 0. Therefore, only one constraint will be active
is to minimize (1/λ) while satisfying the equality constraints, upon the status of the bus.
therefore the objective function to be minimized by the PSO The proposed algorithm was programmed as an M-file in
will be: Matlab. It was tested on two systems, (Ward-Hale 6-bus and
   2 modified IEEE 14-bus systems) to show the applicability of the
1
f (x, λ) = w1 + w2 fpi2 + fqi2 (25) proposed algorithm.
λ i
The initial position of the particles was randomly cho-
where w1 , w2 are two weighting factors for i = 2:N. sen around the power flow solution for some base case.
A.A. EL-Dib et al. / Electric Power Systems Research 76 (2006) 485–492 489

Table 1 Table 3
PSO constants Branch data
w 1.08 → 1.07 Bus code i–m Impedance zim Half of line charging yim /2
a1 1
a2 2 1–4 0.080 + j0.370 0.0 + j0.015
χ 0.92 1–6 0.123 + j0.518 0.0 + j0.021
2–3 0.723 + j1.05 0.0 + j0.0
2–5 0.282 + j0.640 0.0 + j0.0
3–4 0.000 + j0.133 0.0 + j0.0
The constants of the PSO algorithm were chosen as 4–6 0.097 + j0.407 0.0 + j0.015
5–6 0.000 + j0.300 0.0 + j0.0
follows:
As seen in Table 1, the inertia weight (w) decreases linearly
from 1.08 to 1.07 at the end of the iterations. Table 4
These constants gave the best performance throughout the Results of both methods case 1 (pf = 1)
tests. The same constants are used in the algorithm when applied
Bus no. CPF PSO algorithm
on both systems.
V (pu) δ (◦ ) V (pu) δ (◦ )

4.1. Ward-Hale 6-bus system 1 1.05 0 1.05 0


2 0.8777 −17.206 0.8765 −17.2551
3 0.7738 −30.1376 0.7726 −30.2171
The Ward-Hale 6-bus system consists of two generators, four 4 0.7336 −22.9985 0.7325 −23.0528
load buses, and seven branches of which two branches (2–3, 4–5) 5 0.6496 −34.6067 0.6481 −34.7233
are under load tap setting transformers. The system is shown in 6 0.6909 −28.3901 0.6896 −28.4668
Fig. 1. The loads are given in Table 2. The branch parameters λ 0.1596 0.1600
Det (J) 97.3868 0.5176
are given in Table 3. Generator no. 1 is set as the slack bus
with voltage magnitude = 1.05 pu. The second generator is a PV-
bus with scheduled voltage magnitude = 1.1 pu and scheduled
The stopping criterion proposed in this work is either some
generated power = 0.5 pu. The setting of the LTC (2–3) is 0.909,
maximum number of iterations or certain number of stall itera-
while the setting of the one (4–5) is 0.975. The system data is
tions. Another stopping criterion may be some stability margin
given in [17]. The number of particles used in the PSO model
with acceptable power flow equations tolerance. The algorithm
with this system is 80 particles. The penalty factors used were:
solved the problem after 16 s with a tolerance of 4 × 10−4 pu on a
w1 = 0.5, w2 = 105 , w3 = 105 .
PC Pentium 3 400 MHz. The results obtained from the algorithm
were compared to the results from the widely used continuation
power flow (CPF) technique. As shown in Table 4, the results
from both methods for the same conditions are compared. These
conditions are: the load increase is equally distributed among the
PQ-buses with a power factor of unity. This is achieved by set-
ting: kli = 1, ϕi = 0, i.e. Npq .
To prove the effectiveness and robustness of the method
the load increase direction or scenario was changed and the
results were compared with CPF results. As an example, the load
increase power factor was changed to 0.866 lag (let ϕi = π/6).
The results are given in Table 5.
In both cases, the algorithm satisfied also the inequality con-
straint on the reactive power by keeping the reactive power out-
Fig. 1. Ward-Hale 6-bus system.

Table 5
Table 2 Results of both methods case 2 (pf = 0.866)
Operating conditions of 6-bus system
Bus no. CPF PSO algorithm
Bus no. Voltage (pu) Node specification SC (pu)
V (pu) δ (◦ ) V (pu) δ (◦ )
P (pu) Q (pu)
1 1.05 0 1.05 0
1 1.05 – – 2 0.8554 −8.9782 0.8524 −8.9802
2 1.1 0.5 – 0.0 3 0.7552 −24.1158 0.7523 −24.2419
3 −0.55 −0.13 0.0 4 0.7266 −17.7101 0.7242 −17.7819
4 −0.0 −0.0 0.0 5 0.6182 −26.7571 0.6144 −26.9388
5 −0.3 −0.18 0.0 6 0.6832 −22.4714 0.6803 −22.5822
6 −0.5 −0.05 0.0 λ 0.1031 0.1036
Det (J) 173.1013 0.0727
Node 1: the slack bus; SC: shunt capacitors installed at this bus.
490 A.A. EL-Dib et al. / Electric Power Systems Research 76 (2006) 485–492

Table 6 Table 7
Results of both methods (pf = 1, Qmax off) Operating conditions of IEEE 14-bus system
Bus no. CPF PSO algorithm Bus no. Voltage (pu) Node specification SC (pu)

V (pu) δ (◦ ) V (pu) δ (◦ ) P (pu) Q (pu)

1 1.05 0 1.05 0 1 1.06 – –


2 1.1 −42.7885 1.1 −42.9457 2 1.045 0.183 −0.127 0.0
3 0.7827 −41.4363 0.7817 −41.5378 3 1.010 −0.942 −0.19 0.0
4 0.7137 −32.6872 0.7127 −32.7570 4 −0.478 0.039 0.0
5 0.7319 −49.3832 0.7309 −49.5259 5 −0.076 −0.016 0.0
6 0.6779 −39.9867 0.6767 −40.0894 6 1.070 −0.112 −0.075 0.0
λ 0.2502 0.2506 7 0.0 0.0 0.0
8 1.090 0.0 0.0 0.0
9 −0.295 −0.166 0.19
10 −0.090 −0.058 0.0
put equal to the maximum capability of the bus (Qmax = 0.2 pu). 11 −0.035 −0.018 0.0
12 −0.061 −0.016 0.0
Another example is demonstrated where the maximum reactive
13 −0.135 −0.058 0.0
power limit is relaxed as to test the algorithm if it can deal with 14 −0.149 −0.05 0.0
the constraint on the voltage magnitude. The results are shown
in Table 6.
Table 8
Comparison between the results
4.2. IEEE 14-bus system
Bus no. CPF PSO algorithm
The network shown in Fig. 2 has two generators and 12 load V (pu) δ (◦ ) V (pu) δ (◦ )
points. Three of these load points are PV-buses beside one of
1 1.06 0 1.06 0
the generators. The other generator is taken as the slack bus.
2 0.9401 −9.1501 0.9391 −9.1670
The network data is given in [10]. The operating conditions of 3 0.8550 −23.1475 0.8534 −23.2167
the system are shown in Table 7. The algorithm is also able 4 0.8028 −22.4026 0.8008 −22.4711
to solve this system using the same parameters as before. The 5 0.8174 −19.2686 0.8154 −19.3271
number of particles used to solve this system was 160 particles. 6 0.7834 −42.6739 0.7804 −42.9131
7 0.7858 −37.8496 0.7831 −38.045
The penalty factors used were: w1 = 2.5, w2 = 105 , w3 = 105 . The
8 0.8364 −37.8496 0.8338 −38.0427
results obtained compared to the results of the CPF are shown in 9 0.7565 −44.4959 0.7535 −44.7478
Table 8. The load increase scenario is as case 1 in the Ward-Hale 10 0.7386 −46.8278 0.7356 −47.1075
system (kli = 1, ϕi = 0, i.e. Npq ). The algorithm reached the solu- 11 0.7426 −46.9309 0.7396 −47.2152
tion with a tolerance of 4 × 10−4 pu after about 3 min. However, 12 0.7271 −48.627 0.7240 −48.9292
13 0.7271 −48.0311 0.7239 −48.3208
14 0.6928 −51.4631 0.6895 −51.7877
λ 0.1968 0.1975
Det (J) 3.2797 × 1020 1.9643 × 1018

if the tolerance required is 7 × 10−4 the algorithm takes about


2 min on a PC Pentium 3 400 MHz.

5. Analysis of the results

The proposed algorithm was successful in solving the


optimization problem of finding the maximum loading of a
power system subject to some constraints. As shown in the
Tables 4 and 8, the algorithm reached a point very close to
the maximum loading point with an acceptable tolerance. The
convergence characteristic of the PSO algorithm in solving the
problem is shown in Fig. 3. The determinant of the Jacobian
indicates that the algorithm reached a point closer to the max-
imum loading point than the CPF technique. The advantage of
solving this problem as an optimization problem is that any kind
of operational constraints required can be included in the prob-
lem easily. These constraints may include limits on the voltages
Fig. 2. IEEE 14-bus system. of PQ-buses or maximum current flow in the lines. Using the
A.A. EL-Dib et al. / Electric Power Systems Research 76 (2006) 485–492 491

Fig. 3. Convergence c/cs of PSO.

PSO in solving the problem is an advantage since it can deal or integer. It is also capable of finding the global optimum of the
with all of these constraints with the same way shown in the objective function not some local optimum point.
paper by using the penalty function approach. The penalty func-
tion approach may require some effort in choosing the penalty References
factors until acceptable performance is reached.
[1] H.D. Chiang, R. Jean-Jumeau, A more efficient formulation for compu-
tation of the maximum loading points in electric power systems, IEEE
6. Conclusion Trans. Power Systems 10 (May (2)) (1995) 635–646.
[2] M.M. Begovic, A.G. Phadke, Control of voltage stability using sen-
A new application for particle swarm optimization has been sitivity analysis, IEEE Trans. Power Systems 7 (February (1)) (1992)
114–123.
developed. The PSO algorithm has been strengthened using [3] B. Gao, G.K. Morison, P. Kundur, Voltage stability evaluation using
breeding technique similar to that applied in Genetic algorithm modal analysis, IEEE Trans. Power Systems 7 (November (4)) (1992)
(GA). The new suggested algorithm has been applied to two test 1529–1542.
systems (Ward-Hale system and the IEEE 14-bus system). The [4] C.D. Vournas, Voltage stability and controllability indices for multi-
results prove the applicability and validity of the new technique machine power systems, IEEE Trans. Power Systems 10 (August (3))
(1995) 1183–1194.
as a tool for solving the optimization problem of finding the [5] A. Semlyen, B. Giao, W. Janischevskj, Calculation of the extreme
maximum loading of the power system. The advantage of using loading condition of a power system for the assessment of voltage
the PSO algorithm is the ease of formulating and solving the stability, IEEE Trans. Power Systems 6 (February (1)) (1991) 307–
problem with any type of constraints without the need to form 315.
any kind of matrices or getting the inverse of these matrices, [6] O.O. Obadina, G.J. Berg, Determination of voltage stability limit in
multimachine power systems, IEEE Trans. Power Systems 3 (November
which may be in some cases difficult. From the results given, (4)) (1988) 1545–1554.
the PSO algorithm is able to find a solution which is closer to [7] V. Ajjarapu, C. Christy, The continuation power flow: a tool for steady
the maximum point than the CPF technique. state voltage stability analysis, IEEE Trans. Power Systems 7 (February
This HPSO-based algorithm is intended for the planning pur- (1)) (1992) 416–423.
poses. The same algorithm can be modified to include all the [8] G.D. Irisarri, X. Wang, J. Tong, S. Moktari, Maximum loadability of
power systems using interior point non-linear optimization method, IEEE
control variables in the system to increase the maximum load- Trans. Power Systems 12 (February (1)) (1997) 162–172.
ing and hence the stability margin. Here, comes the power of [9] J. Kennedy, R. Eberhart, Particle swarm optimization, Proc. IEEE Int.
PSO where it can easily handle all types of variables either real Conf. Neural Netw. (1995) 1942–1948.
492 A.A. EL-Dib et al. / Electric Power Systems Research 76 (2006) 485–492

[10] Y. Fukuyama, S. Takayama, H. Yoshida, K. Kawata, Y. Nakanishi, A [14] Y. Fukuyama, et al., Practical distribution state estimation using hybrid
particle swarm optimization for reactive power and voltage control con- particle swarm optimization, in: Proceedings of IEEE Power Engineering
sidering voltage security assessment, IEEE Trans. Power Systems 15 Society Winter Meeting, Columbus, 2001.
(November (4)) (2000) 1232–1239. [15] G. Ciuprina, D. Ioan, I. Munteanu, Use of intelligent-particle swarm
[11] Z.-L. Gaing, Particle swarm optimization to solving the economic dis- optimization in electromagnetics, IEEE Trans. Magn. 38 (2002)
patch considering the generator constraints, IEEE Trans. Power Systems 1037–1040.
18 (August (3)) (2003). [16] M. Lovbjerg, T. Kiel Rasmussen, T. Krink, Hybrid particle swarm
[12] A.A. EL-Dib, H.K. Youssef, M.M. EL-Metwally, Z. Osman, Load flow optimizer with breeding and subpopulation, in: Proceedings of
solution using hybrid particle swarm optimization, in: Proceedings of GECCO2001—Genetic and Evolutionary Computation Conference, San
ICEEC’04, Cairo, September, 2004, pp. 742–746. Francisco CA, USA, July 7–11, 2001, p. 409.
[13] Y. Shi, R. Eberhart, Parameter selection in particle swarm optimisation, [17] R.N. Dhar, Computer Aided Power Systems Operation and Analy-
in: Proceedings of Seventh Annual Conference Evolutionary Program, sis, TATA McGraw-Hill Publishing Company Limited, New Delhi,
March, 1998, pp. 591–600. 1982.

You might also like